trainfabric 0.1.13 → 0.1.15

package/dist/index.cjs

@@ -8922,7 +8922,7 @@ function buildComputeSpec(options) {
 
 // src/index.ts
 var DEFAULT_TRAINFABRIC_API_URL2 = "https://api.trainfabric.com";
-var CLI_VERSION = "0.1.13";
+var CLI_VERSION = "0.1.15";
 var CONFIG_DIR = import_node_path3.default.join(import_node_os2.default.homedir(), ".trainfabric");
 var CONFIG_PATH = import_node_path3.default.join(CONFIG_DIR, "config.json");
 var FALLBACK_SECRET_PATH = import_node_path3.default.join(CONFIG_DIR, "session.enc");
@@ -9260,10 +9260,17 @@ function visibleConfig(config) {
     refreshToken: config.refreshToken ? "<stored>" : void 0
   };
 }
-function buildRunInput(options) {
+function requireProjectId(options, config) {
+  const projectId = options.project ?? config.projectId;
+  if (!projectId) {
+    throw new Error("Project is required. Pass --project <projectId> or run `trainfabric config:set-project <projectId>`.");
+  }
+  return projectId;
+}
+function buildRunInput(options, config = loadConfig()) {
   const sourceOptions = buildSourceOptions(options);
   return {
-    projectId: options.project,
+    projectId: requireProjectId(options, config),
     task: "sft",
     method: "lora",
     baseModel: normalizeBaseModel(options.model),
@@ -9285,6 +9292,9 @@ function assertQuoteOptions(bundle, options) {
     return;
   }
   const constraints = [];
+  if (options.mode) {
+    constraints.push(`mode=${options.mode}`);
+  }
   if (options.accelerator) {
     constraints.push(`accelerator=${options.accelerator}`);
   }
@@ -9297,10 +9307,19 @@ function assertQuoteOptions(bundle, options) {
   throw new Error(
     `No quote options were returned for the requested run under the current hardware constraints (${constraints.join(
       ", "
-    )}). Omit --accelerator or increase --min-memory.`
+    )}). Try a different --mode, omit --accelerator, adjust --min-memory if it excludes available GPUs, or retry when capacity changes.`
   );
 }
-async function watchRun(runId, json = false, timeoutMs) {
+function filterQuoteBundleForRequestedMode(bundle, options) {
+  if (!options.mode) {
+    return bundle;
+  }
+  return {
+    ...bundle,
+    quotes: bundle.quotes.filter((item) => item.mode === options.mode || item.quote.mode === options.mode)
+  };
+}
+async function watchRun(runId, json = false, timeoutMs, pollMs) {
   const handle = await createClient(loadConfig()).runs.watch(runId);
   if (json) {
     const emit = (type, payload) => printJson({ type, payload });
@@ -9337,7 +9356,7 @@ async function watchRun(runId, json = false, timeoutMs) {
       console.log(`[failed] ${payload.reason ?? "Run failed."}`);
     });
   }
-  const detail = await handle.wait({ timeoutMs });
+  const detail = await handle.wait({ timeoutMs, pollMs });
   printJson(detail);
 }
 function parseDurationMs(value) {
@@ -9481,36 +9500,43 @@ program2.command("datasets:upload").argument("<file>").option("--project <projec
   });
   printJson(dataset);
 });
-program2.command("runs:create").requiredOption("--project <projectId>").requiredOption("--dataset <datasetId>").requiredOption("--model <baseModel>").option("--eval <evalDatasetId>").option("--epochs <epochs>", "number of epochs", "3").option("--lr <lr>", "learning rate", "0.0002").option("--gpus <gpuCount>", "gpu count", "1").option("--nodes <nodeCount>", "node count", "1").option("--accelerator <acceleratorClass>", "optional hard accelerator constraint (for example: a10g, a100, h100)").option("--min-memory <gigabytes>", "minimum GPU memory in GB").option("--precision <precision>", "fp16_bf16, fp8, or fp32").option("--interconnect <interconnect>", "pcie or nvlink").option("--mode <mode>", "efficient, balanced, or power", "balanced").option("--quote <pricingQuoteId>", "launch the exact accepted quote id from runs:quote").option("--repo <path>", "local repo path for runtime autodetect").option("--git <url>", "git repo URL for runtime metadata").option("--branch <branch>", "git branch for runtime metadata").option("--yes", "confirm that you reviewed pricing with runs:quote --summary and accept fluctuating realized usage").description("Create a training run").action(async (options) => {
+program2.command("runs:create").option("--project <projectId>").requiredOption("--dataset <datasetId>").requiredOption("--model <baseModel>").option("--eval <evalDatasetId>").option("--epochs <epochs>", "number of epochs", "3").option("--lr <lr>", "learning rate", "0.0002").option("--gpus <gpuCount>", "gpu count", "1").option("--nodes <nodeCount>", "node count", "1").option("--accelerator <acceleratorClass>", "optional hard accelerator constraint (for example: a10g, a100, h100)").option("--min-memory <gigabytes>", "minimum GPU memory in GB").option("--precision <precision>", "fp16_bf16, fp8, or fp32").option("--interconnect <interconnect>", "pcie or nvlink").option("--mode <mode>", "efficient, balanced, or power", "balanced").option("--quote <pricingQuoteId>", "launch the exact accepted quote id from runs:quote").option("--repo <path>", "local repo path for runtime autodetect").option("--git <url>", "git repo URL for runtime metadata").option("--branch <branch>", "git branch for runtime metadata").option("--yes", "confirm that you reviewed pricing with runs:quote --summary and accept fluctuating realized usage").description("Create a training run").action(async (options) => {
   if (!options.yes) {
     throw new Error("Refusing to launch without explicit cost acceptance. Run `trainfabric runs:quote --summary ...` first, then rerun `runs:create` with --yes.");
   }
-  const client = createClient(loadConfig());
-  const run = await client.runs.create(buildRunInput(options));
+  const config = loadConfig();
+  const client = createClient(config);
+  const run = await client.runs.create(buildRunInput(options, config));
   printJson(run.snapshot);
 });
-program2.command("runtime:detect").requiredOption("--project <projectId>").option("--repo <path>").option("--git <url>").option("--branch <branch>").description("Detect a supported runtime from a local repo snapshot or git metadata").action(async (options) => {
+program2.command("runtime:detect").option("--project <projectId>").option("--repo <path>").option("--git <url>").option("--branch <branch>").description("Detect a supported runtime from a local repo snapshot or git metadata").action(async (options) => {
   const sourceOptions = buildSourceOptions(options);
   if (!sourceOptions.source) {
     throw new Error("Provide --repo or --git.");
   }
+  const config = loadConfig();
   printJson(
-    await createClient(loadConfig()).runtime.detect({
-      projectId: options.project,
+    await createClient(config).runtime.detect({
+      projectId: requireProjectId(options, config),
       source: sourceOptions.source
     })
   );
 });
-program2.command("runtime:build").requiredOption("--project <projectId>").requiredOption("--detection <detectionId>").description("Build or reuse a cached runtime image from a runtime detection").action(async (options) => {
+program2.command("runtime:build").option("--project <projectId>").requiredOption("--detection <detectionId>").description("Build or reuse a cached runtime image from a runtime detection").action(async (options) => {
+  const config = loadConfig();
   printJson(
-    await createClient(loadConfig()).runtime.build({
-      projectId: options.project,
+    await createClient(config).runtime.build({
+      projectId: requireProjectId(options, config),
       detectionId: options.detection
     })
   );
 });
-program2.command("runs:quote").requiredOption("--project <projectId>").requiredOption("--dataset <datasetId>").requiredOption("--model <baseModel>").option("--eval <evalDatasetId>").option("--gpus <gpuCount>", "gpu count", "1").option("--nodes <nodeCount>", "node count", "1").option("--accelerator <acceleratorClass>", "optional hard accelerator constraint (for example: a10g, a100, h100)").option("--min-memory <gigabytes>", "minimum GPU memory in GB").option("--precision <precision>", "fp16_bf16, fp8, or fp32").option("--interconnect <interconnect>", "pcie or nvlink").option("--mode <mode>", "efficient, balanced, or power").option("--repo <path>").option("--git <url>").option("--branch <branch>").option("--summary", "print a human-readable price summary instead of raw JSON").description("Preview Efficient, Balanced, and Power launch options").action(async (options) => {
-  const bundle = await createClient(loadConfig()).runs.quote(buildRunInput(options));
+program2.command("runs:quote").option("--project <projectId>").requiredOption("--dataset <datasetId>").requiredOption("--model <baseModel>").option("--eval <evalDatasetId>").option("--gpus <gpuCount>", "gpu count", "1").option("--nodes <nodeCount>", "node count", "1").option("--accelerator <acceleratorClass>", "optional hard accelerator constraint (for example: a10g, a100, h100)").option("--min-memory <gigabytes>", "minimum GPU memory in GB").option("--precision <precision>", "fp16_bf16, fp8, or fp32").option("--interconnect <interconnect>", "pcie or nvlink").option("--mode <mode>", "efficient, balanced, or power").option("--repo <path>").option("--git <url>").option("--branch <branch>").option("--summary", "print a human-readable price summary instead of raw JSON").description("Preview Efficient, Balanced, and Power launch options").action(async (options) => {
+  const config = loadConfig();
+  const bundle = filterQuoteBundleForRequestedMode(
+    await createClient(config).runs.quote(buildRunInput(options, config)),
+    options
+  );
   assertQuoteOptions(bundle, options);
   if (options.summary) {
     printQuoteSummary(bundle);
@@ -9535,8 +9561,8 @@ program2.command("runs:wait").argument("<runId>").option("--timeout <duration>",
     })
   );
 });
-program2.command("runs:watch").argument("<runId>").option("--json", "emit event output as JSON").option("--timeout <duration>", "maximum watch duration, for example 30s, 10m, or 1h").description("Stream a run until it completes").action(async (runId, options) => {
-  await watchRun(String(runId), Boolean(options.json), parseDurationMs(options.timeout));
+program2.command("runs:watch").argument("<runId>").option("--json", "emit event output as JSON").option("--timeout <duration>", "maximum watch duration, for example 30s, 10m, or 1h").option("--poll <duration>", "poll interval, for example 1s or 500ms").description("Stream a run until it completes").action(async (runId, options) => {
+  await watchRun(String(runId), Boolean(options.json), parseDurationMs(options.timeout), parsePositiveMs(options.poll, 1e3));
 });
 program2.command("runs:usage").argument("<runId>").option("--summary", "print a human-readable cost summary instead of raw JSON").description("Fetch run usage summary").action(async (runId, options) => {
   const usage = await createClient(loadConfig()).runs.usage(String(runId));

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "trainfabric",
-  "version": "0.1.13",
+  "version": "0.1.15",
   "description": "Trainfabric CLI for launching GPU training jobs on the hosted Trainfabric backend.",
   "type": "module",
   "main": "dist/index.cjs",